Fine-tuning vs. RAG vs. prompting
Three tools, three different problems. Most teams reach for the wrong one first.
Prompting changes instructions. RAG changes knowledge. Fine-tuning changes behaviour. They are not competing options on a scale of sophistication — they solve different failures.
If the model doesn't know your data, fine-tuning will not fix it reliably: fine-tuning teaches form far better than it teaches facts, and it goes stale the moment your data changes. That is RAG's job.
If the model knows the answer but keeps formatting it wrong, breaking your tone, or ignoring a structure, that is prompting first and fine-tuning second — after you have a few thousand good examples and evals to prove it helped.
The shape of it
- 01
Wrong facts
RAG. Give it the source.
- 02
Wrong format or tone
Prompt, then structured output, then fine-tune.
- 03
Too slow or too expensive
Fine-tune a small model on a big model's outputs.
- 04
Wrong task entirely
None of the above. Redesign the feature.
Want this applied to your situation?
Sessions are direct and specific — you leave with a decision, not a reading list.
Book a session